A lightweight WebAssembly runtime that is fast, secure, and standards-compliant
[Cranelift] `(x - y) + y --> x` (#11681)
Bongjun Jang62ce9e5abe08
diff --git a/cranelift/codegen/src/opts/arithmetic.isle b/cranelift/codegen/src/opts/arithmetic.isleindex d4aa06abae..48578730bc 100644--- a/cranelift/codegen/src/opts/arithmetic.isle+++ b/cranelift/codegen/src/opts/arithmetic.isle@@ -331,3 +331,13 @@;; x * (1 << y) == x << y(rule (simplify (imul ty x (ishl ty (iconst_s ty 1) y))) (ishl ty x y))++;; (x - y) + x --> x+(rule (simplify (iadd ty (isub ty x y) y)) x)+(rule (simplify (iadd ty y (isub ty x y))) x)++;; (x + y) - y --> x+(rule (simplify (isub ty (iadd ty x y) x)) y)+(rule (simplify (isub ty (iadd ty x y) y)) x)++diff --git a/cranelift/filetests/filetests/egraph/cprop.clif b/cranelift/filetests/filetests/egraph/cprop.clifindex 37c70ca1b3..87559027a0 100644--- a/cranelift/filetests/filetests/egraph/cprop.clif+++ b/cranelift/filetests/filetests/egraph/cprop.clif@@ -263,8 +263,8 @@ block0(v0: i32, v1: i32):return v7}-; check: v9 = iconst.i32 333-; check: v19 = iadd v1, v9+; check: v15 = iconst.i32 333+; check: v19 = iadd v1, v15; check: v20 = icmp eq v0, v19; nextln: return v20@@ -278,8 +278,8 @@ block0(v0: i32, v1: i32):return v7}-; check: v40 = iconst.i32 333-; check: v49 = iadd v0, v40+; check: v47 = iconst.i32 333+; check: v49 = iadd v0, v47; check: v50 = icmp eq v1, v49; nextln: return v50diff --git a/cranelift/filetests/filetests/egraph/fold-add-sub.clif b/cranelift/filetests/filetests/egraph/fold-add-sub.clifnew file mode 100644index 0000000000..1390faa1df--- /dev/null+++ b/cranelift/filetests/filetests/egraph/fold-add-sub.clif@@ -0,0 +1,60 @@+test optimize precise-output+set opt_level=speed+target x86_64++function %fold_add_sub1(i8, i8) -> i8 {+block0(v0: i8, v1: i8):+ v2 = isub v0, v1+ v3 = iadd v2, v1+ return v3+}++; function %fold_add_sub1(i8, i8) -> i8 fast {+; block0(v0: i8, v1: i8):+; return v0+; }++++function %fold_add_sub2(i16, i16) -> i16 {+block0(v0: i16, v1: i16):+ v2 = isub v0, v1+ v3 = iadd v2, v1+ return v3+}++; function %fold_add_sub2(i16, i16) -> i16 fast {+; block0(v0: i16, v1: i16):+; return v0+; }++++function %fold_add_sub3(i32, i32) -> i32 {+block0(v0: i32, v1: i32):+ v2 = isub v0, v1+ v3 = iadd v2, v1+ return v3+}++; function %fold_add_sub3(i32, i32) -> i32 fast {+; block0(v0: i32, v1: i32):+; return v0+; }++++function %fold_add_sub4(i64, i64) -> i64 {+block0(v0: i64, v1: i64):+ v2 = isub v0, v1+ v3 = iadd v2, v1+ return v3+}++; function %fold_add_sub4(i64, i64) -> i64 fast {+; block0(v0: i64, v1: i64):+; return v0+; }+++diff --git a/cranelift/filetests/filetests/egraph/fold-sub-add.clif b/cranelift/filetests/filetests/egraph/fold-sub-add.clifnew file mode 100644index 0000000000..4bb13bca71--- /dev/null+++ b/cranelift/filetests/filetests/egraph/fold-sub-add.clif@@ -0,0 +1,60 @@+test optimize precise-output+set opt_level=speed+target x86_64++function %fold_sub_add1(i8, i8) -> i8 {+block0(v0: i8, v1: i8):+ v2 = iadd v0, v1+ v3 = isub v2, v1+ return v3+}++; function %fold_sub_add1(i8, i8) -> i8 fast {+; block0(v0: i8, v1: i8):+; return v0+; }++++function %fold_sub_add2(i16, i16) -> i16 {+block0(v0: i16, v1: i16):+ v2 = iadd v0, v1+ v3 = isub v2, v1+ return v3+}++; function %fold_sub_add2(i16, i16) -> i16 fast {+; block0(v0: i16, v1: i16):+; return v0+; }++++function %fold_sub_add3(i32, i32) -> i32 {+block0(v0: i32, v1: i32):+ v2 = iadd v0, v1+ v3 = isub v2, v1+ return v3+}++; function %fold_sub_add3(i32, i32) -> i32 fast {+; block0(v0: i32, v1: i32):+; return v0+; }++++function %fold_sub_add4(i64, i64) -> i64 {+block0(v0: i64, v1: i64):+ v2 = iadd v0, v1+ v3 = isub v2, v1+ return v3+}++; function %fold_sub_add4(i64, i64) -> i64 fast {+; block0(v0: i64, v1: i64):+; return v0+; }+++diff --git a/cranelift/filetests/filetests/runtests/arithmetic.clif b/cranelift/filetests/filetests/runtests/arithmetic.clifindex 47e630c976..9c74c1da19 100644--- a/cranelift/filetests/filetests/runtests/arithmetic.clif+++ b/cranelift/filetests/filetests/runtests/arithmetic.clif@@ -588,3 +588,13 @@ block0(v0: i32, v1: i32):; run: %mul_one_shl(2, 3) == 16; run: %mul_one_shl(4, 5) == 128; run: %mul_one_shl(8, 6) == 512++function %fold_add_sub3(i32, i32) -> i32 {+block0(v0: i32, v1: i32):+ v2 = isub v0, v1+ v3 = iadd v2, v1+ return v3+}++; run: %fold_add_sub3(0xdead0000, 0x0000beef) == 0xdead0000+diff --git a/tests/disas/gc/drc/externref-globals.wat b/tests/disas/gc/drc/externref-globals.watindex b878f09023..c84be143fb 100644--- a/tests/disas/gc/drc/externref-globals.wat+++ b/tests/disas/gc/drc/externref-globals.wat@@ -101,8 +101,8 @@;; @003b v29 = iconst.i64 8;; @003b v14 = iadd v12, v29 ; v29 = 8;; @003b v15 = load.i64 notrap aligned v14-;; v57 = iconst.i64 1-;; @003b v16 = iadd v15, v57 ; v57 = 1+;; v64 = iconst.i64 1+;; @003b v16 = iadd v15, v64 ; v64 = 1;; @003b store notrap aligned v16, v14;; @003b jump block3;;diff --git a/tests/disas/gc/drc/struct-set.wat b/tests/disas/gc/drc/struct-set.watindex 109f0a8379..1a2b233c6b 100644--- a/tests/disas/gc/drc/struct-set.wat+++ b/tests/disas/gc/drc/struct-set.wat@@ -104,8 +104,8 @@;; @004a v33 = iconst.i64 8;; @004a v18 = iadd v16, v33 ; v33 = 8;; @004a v19 = load.i64 notrap aligned v18-;; v62 = iconst.i64 1-;; @004a v20 = iadd v19, v62 ; v62 = 1+;; v69 = iconst.i64 1+;; @004a v20 = iadd v19, v69 ; v69 = 1;; @004a store notrap aligned v20, v18;; @004a jump block3;;